Metrics question

You are a PM of an insurtech aiming to solve the insurance needs of employees within a corporate by trying to launch a voluntary benefits platform. The platform works on providing curated insurance plan combinations rather than pure insurance products. How would you go about developing the plan recommendation system?

Practice this question out loud. An AI interviewer asks it, follows up like a real interviewer would, and scores your answer. Type or speak.

What this question tests

Tests analytical product thinking for a recommendation system: defining inputs, logic, and success metrics for curated insurance bundles.

Clarify the user: an employee choosing voluntary benefits during open enrollment, who is not an insurance expert.
Define the inputs the system needs: employee demographics, existing coverage, dependents, and stated risk tolerance or budget.
Design the recommendation logic: rank curated plan combinations by fit score across cost, coverage gaps, and life stage.
Address trust and transparency: explain why each combination was recommended, not just present a black box ranking.
Define success: percentage of employees who select a recommended bundle versus opting out or choosing manually.
Confirm with the interviewer whether the system launches with a small set of employers or across SoFi's full corporate base.

Grounds inputs in real employee data available at enrollment, dependents, existing coverage, and a short risk questionnaire, rather than assuming rich behavioral data upfront.
Proposes explainable recommendations, showing why a bundle fits, for example flagging a coverage gap for dependents, to build trust in a low familiarity product.
Gives a concrete success metric, for example a stated percentage of employees selecting a recommended bundle over the default opt out.
Addresses cold start explicitly, since new employers have no historical data, by defaulting to demographic based benchmarks.

Designing a purely algorithmic ranking without addressing how employees will trust an unfamiliar bundle recommendation.
Ignoring the cold start problem for employers with no historical enrollment data.

How would you handle an employee who disagrees with the recommendation?
What data would you need from a new corporate client to make this work on day one?